Axios AI+

June 10, 2025
FWIW, Apple quietly announced yesterday afternoon that its forthcoming MacOS 26 Tahoe will be the last major update that will run on Intel-based Macs. Today's AI+ is 1,213 words, a 4.5-minute read.
1 big thing: Apple's missing mojo
Apple's modest AI updates announced yesterday did little to shake the sense that the iPhone maker is still finding its footing in AI as rivals charge ahead.
Why it matters: AI is widely seen as the largest technology shift in decades and could easily serve as an inflection point where existing leaders are dethroned and new ones crowned.
Driving the news: One year after unveiling an expansive vision for personalized AI that it has largely failed to deliver, the iPhone maker focused on a smaller set of tweaks and enhancements to Apple Intelligence at its WWDC conference.
- Some announcements from yesterday, such as live translation, are useful additions already offered on rival devices from Google, Samsung and Microsoft.
- In a handful of other areas, such as image generation, Apple is improving its offering by drawing more heavily on its partnership with OpenAI.
- ChatGPT already handles Apple Intelligence features that require more world knowledge than is available from Apple's smaller, locally run models.
- Apple's most significant AI move was to allow developers to make use of Apple's own models, including a 3 billion-parameter model that runs on Apple devices and a larger one that runs via Apple's servers.
Yes, but: The list of things Apple left unsaid looms larger than the improvements it did announce.
- The company didn't expand — or even really elaborate — on the vision it laid out last year where Apple takes advantage of all that it knows about its individual users to answer questions in a privacy-friendly way.
- Nor did Apple announce rumored deals with Google or Perplexity to serve as additional third-party engines for Apple Intelligence.
- Most glaringly, the company didn't offer a concrete timeline for the improved Siri originally promised last year. Apple's Craig Federighi said only that Apple would have more to say within the coming year.
The big picture: Apple's incrementalism stands in sharp contrast to Google, which unveiled a host of AI features, many of which were the kinds of things that users can touch and use, such as its new tools for video creation.
- Microsoft, Anthropic and others have also held events in recent weeks that offered more substantive advances for developers than what Apple showed yesterday.
- While Apple tends to avoid being at the bleeding edge of technology, its long-standing strategy of arriving late, but polished, might not survive the fast-moving pace of generative AI.
Between the lines: Apple appeared eager not to overpromise at this year's WWDC, announcing only features it expects to be part of the fall release.
- The restraint reflects the fear of repeating last year's disastrous hyping of AI features that slipped past their ship dates.
- However, by sharing only what it is ready to ship, Apple may have reinforced the perception that it has made little progress since last year.
What they're saying: Angelo Zino, a senior vice president and equity analyst at CFRA Research, said he remains positive on Apple for the long term but called yesterday's developer conference a "dud" that is testing investors' patience.
- "In our view, monetization potential for AI is still on the table from both the services and hardware side, but we acknowledge that investors will need to be patient over the next six to nine months as the lack of AI innovation and other concerns linger," Zino wrote.
2. Exclusive: Mistral debuts reasoning models
French AI startup Mistral is launching its first reasoning models, aiming to rival new offerings from OpenAI, DeepSeek and others, Axios has learned.
Why it matters: Reasoning models aim to boost accuracy by using more computing power at inference time, a trend gaining steam amid diminishing returns from building ever-larger language models.
Driving the news: Mistral is releasing two versions of its new Magistral family of reasoning models.
- The first, Magistral Small, is a 24-billion-parameter model that is being released under an open-source license.
- Magistral Medium is more powerful, but proprietary and currently available only in preview mode.
- Mistral says the new models are designed for a range of real-world tasks in areas including law, finance, health care, and engineering. The proprietary version, Mistral says, is capable of 1,000 tokens per second, outpacing the performance of rivals.
Between the lines: Unlike many models that reason primarily in English, Mistral's can "think" in the query's native language, a potentially more efficient approach.
The big picture: Based in France, Mistral has benefited from both its embrace of open source as well as the fact that it offers an alternative to the large American and Chinese tech companies.
- Just 15 months after launching its first paid products, Mistral has racked up over $100 million in contracted sales, according to a source familiar with its finances.
What they're saying: Anjney Midha from a16z, who led the company's Series A funding round and sits on its board, said that Mistral's focus on open source is its key selling point, even more than its unique geography.
- Midha noted that while big shifts in technology tend to be initially led by proprietary efforts, businesses quickly gravitate toward the advantages of open source.
- "They want cheaper, faster and more control," Midha told Axios. "It's happening now with AI. It just takes a little bit of time."
3. Meta launching AI superintelligence lab
Meta CEO Mark Zuckerberg is personally creating a new "superintelligence" team dedicated to building the world's most advanced AI platform, and splashing out nine-figure packages to hire top talent, the New York Times and Bloomberg reported today.
The big picture: Meta is betting that unlimited spending is the key to catching up in the race for AI dominance.
Catch up quick: The Times reported that Scale AI founder Alexandr Wang would join a new Meta lab, designed to develop an AI with powers that ultimately exceed those of the human brain.
- Meta has been in talks to invest billions of dollars in Scale and bring on some of its employees, the Times reported.
- Per Bloomberg, Zuckerberg has been personally leading recruiting for the team of about 50 people — and has even rearranged Meta's offices so the new hires sit near him.
The intrigue: Per the Times, Meta has offered compensation packages between seven and nine figures to AI researchers from top competitors, some of whom have accepted.
- Both the Times and Bloomberg reported that Zuckerberg has been personally frustrated with the pace of Meta's AI development and public stumbles, and has gone into a more hands-on mode to accelerate progress.
- Menlo Ventures venture capitalist Deedy Das posted to X late Monday that he'd heard of at least three instances in the last week of Meta losing out on AI talent to competitors with offers north of $2 million a year.
What to watch: In the highly competitive and lavishly funded industry, such big-ticket efforts don't usually go unanswered.
4. Training data
- OpenAI said it hit $10 billion in annual recurring revenue, driven by consumer and business adoption of ChatGPT. (CNBC)
- A high-stakes copyright trial between Getty Images and Stability AI has kicked off in London's High Court. (Reuters)
- Anthropic killed off an experimental blog written entirely by its chatbot, Claude. (TechCrunch)
- OpenAI says users of ChatGPT, Sora and the OpenAI API are experiencing elevated error rates and the company is working on a fix. (OpenAI)
5. + This
A high school sprinter in Oregon somersaulted her way to victory after tripping on the final hurdle at a recent meet.
Thanks to Scott Rosenberg and Megan Morrone for editing this newsletter and Matt Piper for copy editing.
Sign up for Axios AI+



/2025/06/10/1749516919386.gif?w=3840)

